为解决现有利用左右图像总残差的均值来评价双目立体视觉标定不够精确的问题,提出一种基于对极约束的双目立体视觉标定精度评价方法。该方法充分考虑双目立体视觉中左右图像特征的约束关系及相机标定参数的全局性特征。遵循最小匹配代价原则,该方法利用尺度不变特征变换立体特征匹配法来进行角点检测和匹配。通过左右图像平面上实测角点与其在相对图像平面上对应极线的匹配程度来评价双目立体视觉标定精度,并将这种算法加入到标定算法中,实现了在标定实验过程中对相机标定精度的实时评价。实验表明,该方法比总残差均值法精度更高,精度最高提高了54.0%。

The accuracy of binocular stereo vision calibration obtained using the mean of total residuals of left and right images is unsatisfactory. To overcome this limitation, this study proposes a method for evaluating the accuracy of binocular stereo vision calibration based on the epipolar constraint. The proposed method considers the constraint relationship between left and right image features in binocular stereo vision and global characteristics of camera calibration parameters. Based on the principle of minimum matching cost, a stereo feature matching method based on scale-invariant feature transform is used for corner detection and matching. The accuracy of binocular stereo vision calibration is evaluated based on the matching degree of measured corner points on the left and right image planes with their corresponding epipolar lines on the relative image plane. This proposed algorithm is added to the calibration algorithm to realize real-time evaluation of camera calibration accuracy during the calibration experiment. Experiments demonstrate that the proposed method is more accurate than the method using the mean of residuals, with accuracy increased by up to 54.0%.